1. Gather coverage across the spectrum
We ingest reporting through configured provider adapters (NewsAPI, GNews, or MediaStack), then normalize each item into one shared article shape for downstream processing.
2. Score the source mix
Each source is mapped using a static source rating table and bucketed into left, center, or right. Readers can choose which sources are included, and the resulting source-mix bar updates to match that selection.
3. Synthesize a readable article
The pipeline clusters related coverage for one event, runs deterministic linguistic analysis, and builds a factual core of agreements, contested points, and lean-specific claims before synthesis.
4. Summarise framing differences
Every story includes left, center, and right framing summaries so readers can compare emphasis, omission, and tone without needing to open six browser tabs.
5. Expose the verification trail
Each processed output is canonicalized and signed with Ed25519 during handoff. The published processing file and detached signature can be checked against the public key in the verify workflow.
Example workflow
Topic fetched through provider adapter (or local sample mode) Sources classified with static left / center / right ratings Deterministic language + factual-core stages run on clustered coverage Meta article generated with framing comparison and omissions view Canonical output signed and published with detached verification signature